deformation mapping for coal mining using time-series

Deformation Mapping for Coal Mining using Time-series InSAR Combining

Persistent Scatterers and Distributed Scatterers in Huainan City, China

Zhengjia Zhang1,2，Yixian Tang

1，Chao Wang

1， Hong Zhang

1

1 Key Laboratory of Digital Earth Science, Institute of Remote Sensing and Digital Earth, CAS, Beijing,

China, 100094 2 University of Chinese Academy of Sciences, Beijing, China, 100049

Email:[email protected]

1. INTRODUCTION

Underground coal mining often causes persistent land subsidence, which would lead to huge threats to

infrastructure and safety in the coal area [1, 2]. Traditional levelling and GPS methods are able to produce

reliable measurement of ground subsidence [3]. However, these field surveys are time-consuming and

cannot provide deformation map with high spatial sampling density.

Synthetic aperture radar (SAR) interferometry (InSAR) is a powerful technology, which can obtain

high precise elevation and surface deformation information along line of sight (LOS) using phase

information from different SAR images. However, conventional InSAR usually suffers from temporal,

geometrical and atmospheric decorrelation. In order to overcome these limitations, several advanced

techniques have been proposed, such as Persistent Scatterer Interferometry (PSI) [4, 5] and Small baseline

Subset Algorithm (SBAS) [6]. In recent years, Distributed Scatterers Interferometry (DSI) [7, 8] has been

proposed to monitor the surface deformation with high spatial density of measurement points. The main

drawback of DSI is the computational efficiency [7, 8] in large scale suburb areas.

This paper presents an approach to obtain ground deformation using time-series InSAR technique

combining PSs and DSs. PSs are selected based on their coherence stability in the stack of interferograms.

In order to identify DSs, a selecting strategy combining both classified information and statistical

characteristics is used. To control the error propagation and improve the computational efficiency,

deformation rate and DEM error on PSs are firstly retrieved using conventional PSI. Then a region

grown-based strategy is applied to extract deformation rate on the useful DSs. A series of Radarsat-2 HH

polarization images collected in Huainan are processed to verify the effectiveness of the proposed method.

A dense ground surface deformation map is obtained. The experimental results show that the potential

application of the proposed method.

2. METHOD

The block diagram of the proposed method is shown in Fig. 1. A selecting strategy combining both

classified information and statistical characteristics is used to select DSs. Then the deformation

parameters of PSs are retrieved using conventional PSI. Finally a region growing strategy is applied to

extract deformation rates of DSs. In this study, we focus on two steps: selection of DSs and parameter

estimation of DSs through a region growing method, which will be presented in this section.

2.1 Selection of DSs

DSs can be identified from the combination of classification analysis and mathematical statistics method.

The former is to identify the specific classes for DSs, such as bare soil, ground and the sparse vegetation

areas. The latter is to refine the candidates.

A DS object covers several pixels in high-resolution images, and these pixels are statistical

homogeneous pixels (SHP) sharing the same behaviour corresponding to the suburban and soil area.

However, water and forest area in the image are also homogeneous pixels, but not belonging to DSs. A

prior classification map of the region can be of great help for DSs selection. In this paper, the classification

map is generated by utilizing mean amplitude map of the SAR images to identify potential DSs classes.

For the collected candidates, an adaptive sample selection strategy is applied to refine them. Varies

statistical tests have been proposed in recent years to identify homogenous pixels. The Anderson-Darling

(AD) test has been proven to have better performance than Kolmogorov-Smirnov (KS) test [7]. Thus, AD

test is adopted in this paper. For each candidate in the image, its statistical homogeneous pixels are

identified, which can be applied to adaptive filtering as well as estimation of complex coherence value.

And then, those DS pixels are selected for further processing, which possess an average coherence greater

than a certain threshold and have a minimum number of SHP.

Fig. 1 Block diagram of the proposed method.

2.2 Estimation Differential Deformation rate and DEM error between two points

Generally, the phase difference between two neighbouring pixels, x and y, on each edge of network in the

kth interferogram can be expressed as the following model:

mod ( , ) ( , )

4 4

sin

k k k

el x y x yT v BR

π πϕ ελ λ θ ⊥

∆ = ⋅ ⋅∆ + ⋅ ⋅∆ (1)

Where kT , kB⊥ , R andθ denote time baseline, normal baseline, slant range, and incidence angle, respectively;

λ is the wavelength and v∆ , ε∆ are the deformation rate variation and the elevation error variation of two

neighbouring pixels.

The differential height error and deformation velocity are generated by maximizing the absolute value

of ξ,

mod

1

1exp ( )phase

Nk k

el

k

jN

ξ φ φ=

= ⋅ ∆ − ∆ ∑ (2)

Where N is the number of interferograms, phase

kφ∆ is the phase difference of the neighbouring two points in

the kth interferogram. The maximum of the absolute value of (2) is called temporal coherence.

2.3 Parameters estimation on PSs and DSs

In order to control the error propagation, deformation rate and topographic error on PSs are firstly retrieved

using conventional PSI. After Delaunay network is constructed on PSs, the differential deformation rate

and DEM error on each edge of the network are estimated using the method described in section 2.2. Once

the estimation of deformation parameter of all edges on PSs has been done, the deformation velocity and

DEM error of all the validate points can be obtained through integration step. After getting the linear

deformation rate of PSs, the PSs with good quality are preserved and added into the processed target (PT)

set. The valid deformation results from PS set are used as the reference data for parameters estimation of

DS set.

Because of the large number of the DSs, it will be computationally expensive and difficult to retrieve

deformation parameters if constructi

using region growing strategy is applied

from a reference DS, all the DSs

as unprocessed target (UPT), neighbouring

selecting window (see Fig. 2).

estimated using the method described in section 2.2.

given threshold is viewed as validated link.

viewed as an invalid point. Otherwise, the LOS subsi

using (3) and (4) and the UPT is added into PT set

WhereCV ,

Cε are the estimated deformation rate and DEM error;

iv , i

ε are the deformation rate and DEM error of the i

deformation rate , DEM error, and

In the same way, all the UPT are

Once the linear deformation rate of all the points have been obtained,

components can be extracted using spatial and temporal filtering at both PSs and DSs.

Fig.

Huainan, located in the Yangtze River Delta hinterland

accounted for 19% of the national vision of stocks of coal stock.

mines are located in Huainan City.

which result in many series of environmental problems and damages. As shown in Fig. 3, the central part of

Huainan is selected as the study area, in which there are

XinjiⅠand XinjiⅡ. In the coal mining areas, there are se


deformation parameters if constructing Delaunay triangulation on DSs simultaneously. T

is applied to extract deformation rate and DEM error

all the DSs can be visited through a region growing method. For each DS

neighbouring processed targets (PTs) from the PT set are identified

Then each of the PTs is linked with UPT and v∆ ,∆

using the method described in section 2.2. Only the link with temporal coherence large than a

given threshold is viewed as validated link. If there is no validated links between UPT and PTs

viewed as an invalid point. Otherwise, the LOS subsidence rate and elevation error of the

and the UPT is added into PT set: M

1

1

( )V =

i i iiC M

ii

V vξ

ξ=

=

⋅ + ∆∑∑

M

1

1

( )=

i i iiC M

ii

ξ ε εε

ξ=

=

⋅ + ∆∑∑

are the estimated deformation rate and DEM error; M is the number of the validated links

are the deformation rate and DEM error of the ith PT, respectively ; iv∆ ,

, and temporal coherence of the link between UPT and the

are analysed to retrieve the deformation rate and DEM error.

Once the linear deformation rate of all the points have been obtained, the nonlinear deformation

components can be extracted using spatial and temporal filtering at both PSs and DSs.

Fig. 2 Sketch of identifying processed targets

3. STUDY AREA AND DATA

in the Yangtze River Delta hinterland central of Anhui province, is rich in coal resource,

accounted for 19% of the national vision of stocks of coal stock. About 14 pairs of key

in Huainan City. However, this anthropogenic activity has caused series land subsidence,


e study area, in which there are five coal mining areas, Dingji, Guqiao, Zhangjiaji,

the coal mining areas, there are several collapsed lakes resulting from coal mining.


Therefore, a method

of DS set. Starting

each DS, referring it

are identified using a

v ε∆ of the links are

Only the link with temporal coherence large than a

there is no validated links between UPT and PTs, the UPT is

the UPT are obtained

(3)

(4)

M is the number of the validated links;

v , iε∆ , iξ

denote the

ith PT, respectively.

to retrieve the deformation rate and DEM error.

he nonlinear deformation

central of Anhui province, is rich in coal resource,

s of key state-owned coal

activity has caused series land subsidence,


, Dingji, Guqiao, Zhangjiaji,

resulting from coal mining.

Fig. 3 Location map of the study site and amplitude image. Five

show the location of Anhui province in China and the coverage of a Radarsat

ID Acquisition

Time

Normal

baseline (m)

1 2012/09/24 0

2 2012/12/05 68

3 2013/01/22 198

4 2013/02/15 43

5 2013/03/11 145

6 2013/04/04 155

7 2013/05/22 400

8 2013/06/15 227

9 2013/07/09 257

10 2013/08/02 202

In order to monitor the subsidence of the coal mining area in

images acquired from September

polarization have been used (see Table 1)

all the images for the study area is shown in Fig.3. The pixel spacing of range and azimuth is

2.49 m, respectively. 97 interferometric pairs are combined to generate interferograms with

baseline less than 200 m and temporal baseline less than 200 days.

After generating the average coherence

70939PSs are identified as shown in Fig.

classification map based on the mean amplitude image. By removing the non

candidates are obtained. Subsequently, AD test with a window of 11

which have an average coherence larger than 0.

have been selected by the proposed selection

Fig. 4(B). It is shown that the density of selected DSs is much higher than that of PSs in rural area.

Location map of the study site and amplitude image. Five coal areas are marked. The two insets

show the location of Anhui province in China and the coverage of a Radarsat-

Table 1 Radarsat-2 Data

baseline (m)

Temporal

baseline (days)

ID Acquisition

Time

Normal

baseline (m)

0 11 2013/08/26 191

72 12 2013/09/19 136

120 13 2013/10/13 119

144 14 2013/11/06 183

168 15 2013/11/30 317

192 16 2014/01/17 256

240 17 2014/02/10 306

264 18 2014/03/06 313

288 19 2014/04/23 266

312 20 2014/05/17 253

n order to monitor the subsidence of the coal mining area in Huainan city, 20 Radarsat

acquired from September 2012 to May 2014 with a look angle of 36.6 degrees and

(see Table 1). The SAR amplitude images (about 20×20 km

all the images for the study area is shown in Fig.3. The pixel spacing of range and azimuth is

respectively. 97 interferometric pairs are combined to generate interferograms with

less than 200 m and temporal baseline less than 200 days.

4. EXPERIMENTAL RESULT

the average coherence map, point with coherence value large than 0.6

as shown in Fig. 4(A). The unsupervised IsoData method is used to generate the

classification map based on the mean amplitude image. By removing the non-DSs classes, the DS

Subsequently, AD test with a window of 11 ×11 is applied

which have an average coherence larger than 0.45 and the number of SHP larger than

have been selected by the proposed selection strategy (approximately 2856 DSs/km2), which is shown in

nsity of selected DSs is much higher than that of PSs in rural area.

coal areas are marked. The two insets

-2 image

baseline (m)

Temporal

baseline (days)

336

360

384

408

432

480

504

528

576

600

Radarsat-2 ascending

with a look angle of 36.6 degrees and ‘HH’

20 km2) averaged from

all the images for the study area is shown in Fig.3. The pixel spacing of range and azimuth is 2.66 m and

respectively. 97 interferometric pairs are combined to generate interferograms with perpendicular

65 is selected as PS.

ta method is used to generate the

DSs classes, the DS

is applied to select the DSs

larger than 70. 1,142,592 DSs

, which is shown in

nsity of selected DSs is much higher than that of PSs in rural area.

Experimental result on PSs is shown in Fig.

about 53.3mm/year. It is shown

subsidence rate of 107 mm/year.

which indicates that coal mining contribute to the ground movement.

parameter estimation of DSs and collecting

This paper has presented an approach to extract ground movement in coal mining area using time

InSAR on PSs and DSs, thus resulting in more detail of the subsidence me

efficiently identified using information and

in the processing of DSs, which may make parameter estimation easier and more efficiency

will focus on parameter estimation of DSs.

Fig.4 (A) selected PSs; (B) selected DSs

rimental result on PSs is shown in Fig. 5. The average subsidence velocity of the study area is

It is shown that four subsidence areas are marked by red rectangle

. Those subsidence areas are located in the central of the coal mining area,

which indicates that coal mining contribute to the ground movement. Future in-depth research will focus on

parameter estimation of DSs and collecting ground truth data.

Fig.5 Subsidence velocity map on PSs.

5. CONCLUSION

This paper has presented an approach to extract ground movement in coal mining area using time

, thus resulting in more detail of the subsidence measurements

efficiently identified using information and statistical property. A region growing strategy has been applied

in the processing of DSs, which may make parameter estimation easier and more efficiency

will focus on parameter estimation of DSs.

he average subsidence velocity of the study area is

rectangle line with a peak

hose subsidence areas are located in the central of the coal mining area,

h research will focus on

This paper has presented an approach to extract ground movement in coal mining area using time-series

ments. DSs can be

region growing strategy has been applied

in the processing of DSs, which may make parameter estimation easier and more efficiency. Future work

ACKNOWLEDGMENTS

This work was supported by the National Natural Science Foundation of China under Grants 41271425 and

41331176.

6. REFERENCE

[1] H. C. Jung, S.-W. Kim, H.-S. Jung, K. D. Min, and J.-S. Won, “Satellite observation of coal mining subsidence by

persistent scatterer analysis,” Eng. Geol., vol. 92, pp. 1-13, 2007.

[2] J. Baek, S.-W. Kim, H.-J. Park, H.-S. Jung, K.-D. Kim, and J. W. Kim, “Analysis of ground subsidence in coal

mining area using SAR interferometry,” Geosci. J., vol. 12, pp. 277-284, 2008.

[3] A. Demoulin, J. Campbell, A. D. Wulf, A. Muls, R. Arnould, B. Görres, et al., “GPS monitoring of vertical ground

motion in northern Ardenne–Eifel: five campaigns (1999–2003) of the HARD project,” Int. J. Earth. Sci., vol. 94, pp.

515-524, 2005.

[4] A. Ferretti, C. Prati, and F. Rocca, “Permanent scatterers in SAR interferometry,” IEEE Trans. Geosci. Remote Sens.,

vol. 39, pp. 8-20, 2001.

[5] A. Hooper, H. Zebker, P. Segall, and B.Kampes, “A new method for measuring deformation on volcanoes and other

natural terrains using InSAR persistent scatterers,” Geophys. Res.Lett., vol. 31, 2004.

[6] P. Berardino, G. Fornaro, R. Lanari, and E.Sansosti, “A new algorithm for surface deformation monitoring based on

small baseline differential SAR interferograms,” IEEE Trans. Geosci. Remote Sens., vol. 40, pp. 2375-2383, 2002.

[7] A. Ferretti, A. Fumagalli, F. Novali, C. Prati, F. Rocca, and A. Rucci, “A new algorithm for pro-cessing

interferometric data-stacks: Squeesar,” IEEE Trans. Geosci. Remote Sens., vol. 49, pp. 3460-3470, 2011.

[8] K. Goel and N. Adam, "An advanced algorithm for deformation estimation in non-urban areas," ISPRS J.

Photogram. Remote Sens., vol. 73, pp. 100-110, 2012.

[10] A. Parizzi, and R. Brcic, “Adaptive InSAR stack multilooking exploiting amplitude statistics: a comparison

between different techniques and practical results,” IEEE Trans. Geosci. Remote Sens., vol. 8, pp.441-445, 2011.

deformation mapping for coal mining using time-series

Documents